Talknotes: 17 Built, 1 Hit — the Breakdown Behind an AI Voice-Notes App at $9,000/Month and 900 Paying Users

Dollar figures below are paired with rough yen equivalents at roughly 1 USD = 150 JPY.

17 built, 1 hit

Talknotes is an app that uses AI to turn recorded voice memos into blog posts or bullet-point lists. It was built by Nicolas Jeanne (known as Nico), based in Bangkok, then 26 years old. No employees.

Launch was August 2023. Core feature development took 3-4 days, and upfront cost was under $10 (about ¥1,500), just the OpenAI API usage fee. Current user count is 5,000, with over 900 paying. Total monthly revenue is around $9,000 (about ¥1.35M), broken down into roughly $4,000 in MRR plus the rest as amortized annual contracts.

What’s notable is the number of failures behind this one hit. In his first year, he shipped 17 products, and one succeeded. Counting further back, he’s racked up 30 failures over seven years. “If it’s not working, ship fast and move to the next one” is how he puts it.

The path to launch

For reference, he’d previously sold a different app for $65,000 (about ¥9.75M). The $150,000 (about ¥22.5M) offer for Talknotes came in at just $3,000 MRR.

What it actually sells

Voice-to-text transcription itself is no longer rare. What Talknotes sells is the step after that. The formatting. It takes what you spoke and reshapes it into a blog-post format, a bullet list, or a purpose-specific format. It fully takes over the editing step that sits between “record it” and “get usable text.”

The pricing design changed over time. Initially, annual lump-sum payment was the core offer, with an on-ramp from a limited free version. A monthly subscription was added later. What matters here is that the annual plan carried a steep discount, the reasoning behind that comes below.

The technical stack is unpretentious: originally vanilla JavaScript, HTML, and CSS, recently migrated to Nuxt. Coding uses Cursor, ChatGPT, and Copilot. Analytics run on Plausible and Simple Analytics, email on Bento, payments on Stripe, hosting on Heroku, a standard indie-dev setup.

The decisive call: feeding annual prepayments straight into ads

What set Talknotes’ trajectory was the choice of acquisition channel, and the cash-flow design that supported it.

Most indie developers head to SEO first. Nico tried it too. The result: only 3 sales came from blog posts. Free-traffic tools performed no better. He pivoted from there and shifted fully toward paid advertising, Facebook, Instagram, Google, Twitter, LinkedIn, Bing, essentially every platform he could run ads on. Customer acquisition cost (CAC) came out to about $50 (about ¥7,500).

The question was whether a solo developer could sustain the cash flow to keep buying ads. What made it work was the steep discount on the annual plan. Each annual signup brought in 12 months of cash upfront. He fed that prepaid cash straight back into ad spend, not waiting for monthly profit to accumulate, but funding next month’s ad budget with this month’s annual signups. As long as an annual plan’s per-unit value exceeded the $50 CAC, the loop kept running on its own.

In numbers: monthly growth rate around 10%, churn under 10%. Today’s 900+ paying customers is the result of running this loop continuously. Recall that in week one post-launch, only 7 of 1,000+ free users converted to an annual contract. That context underscores the weight of his decision not to chase free-user counts.

Why this loop keeps turning

Three conditions mesh together here.

First, the product’s value is instantly obvious. Ads give you a short window to explain. A product that can be summed up in one line (“just talk, and it becomes text”) is what makes a $50 CAC viable at all. Flip it around: run the same tactic with a product that needs a longer explanation, and CAC would blow up immediately.

Second, the annual discount stands in for financing. Running ads without external capital means using customer prepayments as the funding source instead. A discount technically erodes lifetime value, but it makes sense once you view it as the cost of buying growth speed. There’s a clear tradeoff here, and he’s choosing speed with eyes open.

Third, churn is low. Sub-10% monthly churn shows that customers acquired via ads are actually sticking around. If customers acquired via paid CAC left immediately, this loop would collapse on its first cycle. Acquisition and product retention aren’t separable questions here.

As a side product, he also built a “Facebook ads guide for SaaS,” earning $25,000 (about ¥3.75M) on presale alone, turning his own ad-running experience into a secondary product.

What didn’t work, and the risks he’s carrying

SEO clearly failed. Blog posts produced only 3 sales. None of the tools aimed at capturing free traffic paid off either. For a freshly launched solo product, the months-to-years timeline for search rankings to mature likely wasn’t a realistic fit.

Even so, he currently lists SEO again as a “priority going forward.” The reason is obvious: revenue is almost 100% dependent on paid advertising. A single event (rising ad platform costs, an account suspension, a change in review standards) could halt new acquisition entirely. Bringing down CAC and building out a channel outside of paid ads are the next challenges he names himself.

Another risk is competitive density. Voice input combined with AI summarization has a low barrier to entry, and could easily be absorbed into an OS-native feature. The $150,000 offer at just $3,000 MRR is both a signal of outside valuation and evidence of just how liquid a market a business at this scale sits in.

The personal cost is also on record. He says the stress of building the business gave him gray hair. There’s nothing flat about what sits behind these numbers.

What can be copied, what can’t

What’s transplantable is concrete: shipping core functionality in days to test it against the market. The decision to cut losses quickly once free traffic proves not to work. And the cash cycle of feeding annual prepayments into ad spend. For a capital-constrained individual, these three offer realistic options to consider.

But the preconditions need to be looked at directly too. He’d previously sold an app for $65,000, and already had both the capital and the know-how to experiment with running ads. Attempting to launch across six ad platforms and optimize them from a true zero start would be far harder. Also worth remembering: this is 1 hit out of 17 attempts. What appears in this article is the one that landed, the time and money spent on the other 16 don’t show up in these numbers.

And the biggest factor that can’t be replicated is timing. August 2023 was a moment when small apps embedding generative AI were still perceived as “new,” able to draw attention on Product Hunt. Launching the same product today wouldn’t get the same initial velocity. What can actually be extracted from Talknotes’ numbers isn’t the reason it won. It’s the structure of the loop that won.

Related reading

• PhotoAI $132K/month, zero employees, 37,000 commits a year. Dissecting Pieter Levels’ “run it solo” management style
• 30th indie project “MENTA”: from ¥1.4M/month to an equity transfer to Lancers. Shingo Irie’s full story